Size and distribution of the global volume of surgery in 2012

Abstract Objective To estimate global surgical volume in 2012 and compare it with estimates from 2004. Methods For the 194 Member States of the World Health Organization, we searched PubMed for studies and contacted key informants for reports on surgical volumes between 2005 and 2012. We obtained data on population and total health expenditure per capita for 2012 and categorized Member States as very-low, low, middle and high expenditure. Data on caesarean delivery were obtained from validated statistical reports. For Member States without recorded surgical data, we estimated volumes by multiple imputation using data on total health expenditure. We estimated caesarean deliveries as a proportion of all surgery. Findings We identified 66 Member States reporting surgical data. We estimated that 312.9 million operations (95% confidence interval, CI: 266.2–359.5) took place in 2012, an increase from the 2004 estimate of 226.4 million operations. Only 6.3% (95% CI: 1.7–22.9) and 23.1% (95% CI: 14.8–36.7) of operations took place in very-low- and low-expenditure Member States representing 36.8% (2573 million people) and 34.2% (2393 million people) of the global population of 7001 million people, respectively. Caesarean deliveries comprised 29.6% (5.8/19.6 million operations; 95% CI: 9.7–91.7) of the total surgical volume in very-low-expenditure Member States, but only 2.7% (5.1/187.0 million operations; 95% CI: 2.2–3.4) in high-expenditure Member States. Conclusion Surgical volume is large and growing, with caesarean delivery comprising nearly a third of operations in most resource-poor settings. Nonetheless, there remains disparity in the provision of surgical services globally.


Introduction
Surgical care is essential for managing diverse health conditions -such as injuries, obstructed labour, malignancy, infections and cardiovascular disease -and an indispensable component of a functioning health system. [1][2][3] International organizations, including the World Health Organization (WHO) and the World Bank, have highlighted surgery as an important component for global health development. 3,4 However, surgical care requires coordination of skilled human resources, specialized supplies and infrastructure.
As low-and middle-income countries expand their economies and basic public health improves, noncommunicable diseases and injuries comprise a growing proportion of the disease burden. 5 Investments in health-care systems have increased in the last decade, but the effect on surgical capacity is mostly unknown. 6,7 Based on modelling of available data, it was estimated that 234.2 million operations were performed worldwide in 2004. 8 The majority of these procedures took place in high-income countries (58.9%; 138.0 million), despite their relative lower share of the global population.
Here, we estimated the global volume of surgery in 2012. We also estimated the proportion of surgery due to caesarean delivery, since studies done in low-income countries have found that emergency obstetric procedures -especially caesarean deliveries -represent a high proportion of the total surgical volume. 9,10

Methods Population and health databases
For the years 2005 to 2012, we obtained population and health data for 194 WHO Member States. These data included total population, life expectancy at birth, percentage of total urban population, gross domestic product (GDP) per capita in United States dollars (US$) and total health expenditure per capita in US$. 6,11 For 11 Member States, where certain population or health data were not available from either WHO or the World Bank, we used data from other similar sources. 12,13 All US$ were adjusted for inflation to the year 2012, using the consumer price index for general inflation. 14 For Member States with reported surgical data, we also obtained population and health data from the year for which surgical volume was reported. We classified Member States based on their health spending. Member States spending US$ 0-100 per capita on health were classified as very-low-expenditure Member States (n = 50); US $101-400 as low-expenditure Member States (n = 54); US$ 401-1000 as middle-expenditure Member States (n = 46); and over US$ 1000 as high-expenditure Member States (n = 44). 8

Surgical data sources
Operations were defined as procedures performed in operating theatres that require general or regional anaesthesia or profound sedation to control pain. We searched PubMed for the most recent annual surgical volume reported after 2004, using each Member State name along with the following keywords and phrases for all WHO Member States: "surgery", "proce-Objective To estimate global surgical volume in 2012 and compare it with estimates from 2004. Methods For the 194 Member States of the World Health Organization, we searched PubMed for studies and contacted key informants for reports on surgical volumes between 2005 and 2012. We obtained data on population and total health expenditure per capita for 2012 and categorized Member States as very-low, low, middle and high expenditure. Data on caesarean delivery were obtained from validated statistical reports. For Member States without recorded surgical data, we estimated volumes by multiple imputation using data on total health expenditure. We estimated caesarean deliveries as a proportion of all surgery. Findings We identified 66 Member States reporting surgical data. We estimated that 312.9 million operations (95% confidence interval, CI: 266.2-359.5) took place in 2012, an increase from the 2004 estimate of 226.4 million operations. Only 6.3% (95% CI: 1.7-22.9) and 23.1% (95% CI: 14.8-36.7) of operations took place in very-low-and low-expenditure Member States representing 36.8% (2573 million people) and Global volume of surgery Thomas G Weiser et al.
dures", "operations", "national surgical volume" and "national surgical rate". Depending on the Member State, we conducted our search in English, French and/or Spanish. To obtain email addresses for ministers or officials working for the ministry of health or individuals responsible for auditing surgical data at a national level, we searched the internet for the websites of ministries of health or national statistical offices. We contacted these persons to request the most recently reported total volume of operations based on the above definition. From the database of the Organisation for Economic Co-operation and Development (OECD) we obtained surgical volume for 26 countries; 14 of these countries had total surgical volume data as well as detailed data for a subset of procedures (termed a shortlist by OECD), while the other 12 countries only had data for the shortlist. 15 For the 14 countries, we used both data sets in combination with publicly available data on total health expenditure to define the relationship between the shortlist and the reported total surgical volume. We used this relationship to estimate total surgical volume for the 12 countries that only had shortlist and total health expenditure data. The average relative difference between the observed total surgical rate and extrapolated total surgical rate was 13.7% for these 14 countries; in a leave-one-out cross validation, the relative average bias was 16%.
For the Member States from which we obtained surgical data between 2005 and 2013, we calculated the annual surgical volume per 100 000 population for the year that the data were reported for the Member State by using the total population estimate for the same year.

Model development
To develop a predictive model for surgical rates, we first investigated the bivariate Spearman correlations between surgical rate and five a priori countrylevel variables: total population, life expectancy, percent urbanization, GDP per capita and total health expenditure per capita. We selected total health expenditure per capita as the only explanatory variable based on the results of Spearman correlations. We then did two sensitivity analyses: Spearman partial correlations and a multivariable regression model using the Lasso approach for variable selection. 16 Our final predictive model contained only total health expenditure per capita. Finally, we log-transformed total health expenditure per capita and surgical rate to account for their rightskewed distribution.

Missing data analysis
To determine if any of the five a priori country-level predictors was related to the probability that a country's surgical rate was missing, we fitted a multivariable logistic regression (Table 1). 17 This model allowed us to determine variables associated with surgical rate. These variables could then be included in the imputation model to predict the rates for the Member States with missing data. The only variable significantly associated with whether a country's surgical rate was missing was total health expenditure per capita, which was already included in the imputation model.

Imputation model
To find the best fitting model for the relation between surgical rate and total per capita health expenditure, we built a spline model, positing splines with zero, one, two or three inflection points. [18][19][20] The best-fitting spline model was selected based on leave-one-out cross-validation, in which the predicted surgical rate value for a country was estimated based on a model that had been fitted after omitting data for that country. We used total per capita health expenditure from 2012 for our imputation model of surgical rates. The Democratic People's Republic of Korea, Somalia and Zimbabwe had no available total health expenditure data for 2012. Since the Pearson correlation between health expenditure in 2012 and any single year between 2000 and 2011 for all other Member States was ≥ 0.97, we extrapolated total health expenditure for these Member States by using their expenditure from previous years. As we did not have reported total health expenditure for 2013, we assumed that surgical rates or volume reported for 2013 were equivalent to 2012 values. For the 25 Member States with surgical data reported before 2012, we extrapolated 2012 estimates for these using a multiple imputation model that treated 2012 For Member States with missing surgical volume data, we used multiple imputation and our predictive model to arrive at 2012 surgical rate estimates. 21 We produced 300 imputed data sets to estimate the mean global surgical volume and its corresponding 95% confidence interval. Using the imputed country-level surgical rates and population estimates for 2012 we calculated the number of operations performed in each country in 2012. We also used published caesarean delivery data to calculate the proportion of surgical volume accounted for by caesarean delivery for each country. 22 These data came primarily from the Global Health Observatory data repository, 23 World Health Statistics 2010, 24 the World Health Report 2010, 25 the Demographic and Health Surveys 26 and OECD. 15 To compare the 2004 estimates with the new 2012 estimates, we used the same data on reported surgical rate from 56 countries that we used in the 2004 modelling exercise 8 and did a spline analysis. We tested spline models with zero, one, two or three inflection points for the 2004 data. The spline model with two inflection points had the highest adjusted cross validation R 2 , as with the 2012 data. We evaluated the change in surgical rates that occurred for each health expenditure group between 2004 and 2012. This ensured that any observed changes in estimated volume were not driven by the updated modelling approach (details available from corresponding author).
We used SAS software version 9.2 (SAS Institute Inc., Cary, United States of America) for all statistical analyses. Two-sided statistical tests were done and all P-values less than 0.05 were considered statistically significant.

Model development
The total health expenditure per capita was the most highly correlated variable with surgical rate (Spearman correlation, r = 0.87297; P < 0.0001; Table 2; available at: http://www.who.int/bulwletin/volumes/94/3/15-159293). The sensitivity analyses showed that after adjusting for total health expenditure per capita, none of the other variables remained significant. WHO regions were also not significantly associated with surgical rate (P = 0.09). Fig. 1 shows the best fitting spline model for surgical rate based on total health expenditures, with two inflection points at US$ 288 and US$ 1950 per person per year (r 2 : 0.7449). The models with zero, one and three inflection points had adjusted cross validation r 2 of 0.7064, 0.7071 and 0.7332 respectively.

Surgical volume
We obtained surgical data from 66 Member States (Table 3; available at: http://www.who.int/bulletin/vol/umes/94/3/15-159293). Using multiple imputation, we extrapolated the volume of surgery for each country without reported surgical data ( Caesarean delivery data were more widely available than overall surgical data, with data from 172 Member States. In very-low-expenditure settings, caesarean delivery accounted for 29.6% (5.8/19.6 million operations) of all operations performed. However, in high-expenditure Member States this percentage was only 2.7% (5.1/187.0 million operations; Table 6). Worldwide, caesarean deliveries account for nearly one in every 14 operations performed.

Discussion
We estimate 266.2 to 359.5 million operations were performed in 2012. Global volume of surgery Thomas G Weiser et al.
This represents an increase of 38% over the previous eight years. We note the largest increase in operations was in very-low-and low-expenditure Member States. However, about one in every 20 operations globally was done in verylow-expenditure Member States, despite these Member States representing well over one third of the total global population. Comparing very-low-expenditure Member States with high-expenditure Member States, the gap in access is even larger. These disparities may be even larger when examining the distribution of access to surgical care within individual Member States, an undertaking that is beyond the scope of this study. The proportion of caesarean delivery were higher in Member States with lower surgical volume. This likely demonstrates that obstetrical emergencies are prioritized as a surgical intervention in Member States with scarce resources, but also suggests that other surgical conditions are left poorly attended in these settings. The findings serve to highlight the importance of improving surgical capacity to address both obstetrical and other surgical conditions.
Surgical data were lacking from many Member States. Compared with the data availability for the 2004 estimates, only 10 more Member States now had available data. This contrasted with caesarean delivery data, which were available for the majority of Member States. Given the efforts of the maternal health community and the importance of caesarean delivery in supporting improved maternal outcomes, our findings are not surprising. The challenge of accessing data on surgical care impede the understanding and monitoring of surgery as a component of global health care. Without standardized and accessible data, it is difficult for researchers and policy-makers to contextualize and prioritize surgical access and quality when discussing health system strengthening.
In 2015, the World Health Assembly passed a resolution strengthening emergency and essential surgical care and anaesthesia as a component of universal health coverage. 72 The increases in injuries and noncommunicable diseases present a challenge for weak health systems already struggling with a high infectious burden of disease. 73 Not only do injuries and many noncommunicable diseases require surgical intervention, in many resource-poor settings neglected infections -such as typhoid and tuber- Table 5.

Research
Global volume of surgery Thomas G Weiser et al.
culosis -are not treated in a timely fashion and therefore require surgical care. 74 The increase in surgical output in very-low-expenditure Member States over the last eight years suggests that these Member States are placing an increasing importance on access to emergency and essential surgical services. However, the Lancet Commission on Global Surgery has estimated that five billion people lack access to safe, affordable surgical and anaesthesia care when needed and an additional 143 million operations are required to address emergency and essential conditions in low-and middle-income countries. 3 The lack of standardized surgical data globally is both a limitation of and the reason for undertaking this study. As part of the WHO Safe Surgery Saves Lives programme for which the 2004 estimates of global surgical volume was performed, our group proposed a standard set of metrics for surgical surveillance. 75 We continued to have difficulty during this study obtaining standardized data regarding surgical intervention. The data were not located or reported in any standardized way and required our research team to compile the information from multiple agencies, ministries, health reports and published literature, as there was no central source for collecting or reporting these data. Some ministry reports may include only state and government facilities and not hospitals run privately or by nongovernmental organizations, which can provide substantial surgical capacity. Thus the volume we report may be an underestimate. Regardless, the non-included facilities are unlikely to close the gap in care between Member States or change our findings. In addition, there was no differentiation between surgical care undertaken in urban versus rural areas. There is likely a large discrepancy in surgical access and provision of surgical care within a single country.
OECD, which had previously collected total operative volume as reported in our last study, 8 has changed its methods and now reports on only a subset of procedures. Thus our analysis required an additional step to turn these data into comprehensive estimates of volume, adding another layer of uncertainty.
Many of the same limitations of the previous analysis were present here. We focused on operations performed in an operating theatre as these are most likely to involve high complexity, acuity and risk. Our study is thus limited by the manner in which such operations and procedures are recorded. We recognize that many minimally invasive procedures can be undertaken outside an operating theatre, as can many imageguided procedures, thus potentially undercounting what might be considered surgery in these settings. Many minor procedures may also be undertaken in the operating room to improve pain control or exposure or because of availability of resources and equipment, thus creating variability within our count. However, by standardizing our definition, we limited the difficulties associated with the variability in case mix and practice patterns across Member States and settings.
As only one third of Member States reported data on surgical volume, our estimates of overall volume of surgery continue to rely on modelling techniques. We noted changes in the slope of the curve of our spline regression over the range of health expenditure, in particular between the two spline inflections, likely reflecting the heterogeneity of Member States. Furthermore, while the imputation strategy was aimed at a global estimate, the estimate for any particular country may be imprecise. However, our modelling strategy was based on the strong explanatory power of per capita expenditure on health as a determinant of surgical volume. Health expenditure per capita was the only variable that was significantly associated with whether surgical rate data was missing, and multiple imputation protects against systemic bias from data that are missing at random.

Conclusion
Surgical volume continues to grow, particularly in very-low-and lowexpenditure Member States. However, surgical surveillance continues to be weak and poorly standardized and limits the precision of these estimates, yet the systematic evaluation of access, capacity, delivery and safety of care is paramount if surgical services are to support a programme of health system strengthening. Furthermore, the relationship of surgical provision to population health outcomes is not clear, and interventions such as surgery that include substantial risk to patients must be carefully considered. Many patients receive surgical care, yet safety and quality-of-care remain poorly measured and a low priority in many Member States. ■
Global volume of surgery Thomas G Weiser et al.

Research
Global volume of surgery Thomas G Weiser et al.